1. Field
The following description relates to a method for motion estimation and an apparatus for processing an image.
2. Discussion of Related Art
Motion estimation is the process of estimating how an object moves between time-series frames when a display device displays a dynamic image, such as a moving image. Motion estimation is used in various apparatuses for dynamic image processing, such as moving image compression of the Motion Picture Experts Group (MPEG), etc., three-dimensional (3D) noise removal, and frame rate conversion.
There is one existing motion estimation method of dividing a reference image into macroblocks having a uniform size, setting a search region for each macro block in a preceding or following frame, and then calculating the sum of inter-pixel luminance differences between a macroblock of the reference image and the corresponding macroblock of the preceding or following frame according to the macroblocks to find a block having the smallest sum. The sum of inter-pixel luminance differences is calculated using a sum of absolute difference (SAD) given by Equation 1.
                                                        S              ⁢                                                          ⁢              A              ⁢                                                          ⁢                              D                ⁡                                  (                                      x                    ,                    y                                    )                                                      =                                          ∑                                  i                  ,                                      j                    ∈                    W                                                              ⁢                                                                                    I                    ⁡                                          (                                              i                        ,                        j                        ,                        t                                            )                                                        -                                      I                    ⁡                                          (                                                                        i                          +                          x                                                ,                                                  j                          +                          y                                                ,                                                  t                          +                          1                                                                    )                                                                                                                    ,          or                ⁢                                  ⁢                              S            ⁢                                                  ⁢            A            ⁢                                                  ⁢                          D              ⁡                              (                                  x                  ,                  y                                )                                              =                                    ∑                              i                ,                                  j                  ∈                  W                                                      ⁢                                                                          I                  ⁡                                      (                                          i                      ,                      j                      ,                      t                                        )                                                  -                                  I                  ⁡                                      (                                                                  i                        +                        x                                            ,                                              j                        +                        y                                            ,                                              t                        -                        1                                                              )                                                                                                                        [                  Equation          ⁢                                          ⁢          1                ]            
Since blocks in a moving object have motion vectors that are similar to each other, it is also possible to use a smoothness constraint (SC) method of calculating differences between a candidate motion vector in a search region of a target image and motion vectors of blocks around a macroblock of a reference image on which motion estimation is performed. An SC value may be calculated as given by Equation 2 below.
                              S          ⁢                                          ⁢                      C            ⁡                          (                              x                ,                y                            )                                      =                              ∑            i                    ⁢                                                                (                                  x                  ,                  y                                )                            -                              v                i                                                                                    [                  Equation          ⁢                                          ⁢          2                ]                                          cost          ⁡                      (                          x              ,              y                        )                          =                              S            ⁢                                                  ⁢            A            ⁢                                                  ⁢                          D              ⁡                              (                                  x                  ,                  y                                )                                              +                      λ            ·                          SC              ⁡                              (                                  x                  ,                  y                                )                                                                        [                  Equation          ⁢                                          ⁢          3                ]            
As given by Equation 3 above, it is possible to perform motion estimation using a cost function that is obtained by multiplying a constant and an SC term to uniformize the weight of an SAD term and the weight of the SC term with respect to a candidate vector (x, y) and summing the SAD term and the SC term multiplied by the constant.
However, when an object includes a region in which regular designs are shown and moves during two successive frames, the regular designs move along with the movement of the object. Upon motion estimation of the object, a pixel difference becomes as small as possible in many blocks due to the regular designs. Therefore, a region in which the object is estimated to move because the cost function is calculated to be the smallest value differs from a region in which the object actually moves. For this reason, errors frequently occur in the motion estimation, and image quality is degraded.
To solve this problem according to related art, motion estimation and pattern recognition are separately performed, and an image is generated by combining two results thereof to perform image processing, such as interpolation. However, when motion estimation and pattern recognition are separately performed as mentioned above, motion information of an object including a region in which regular designs are shown is calculated incorrectly, and pattern recognition and image processing are performed based on the incorrectly calculated motion information of the object. Consequently, the quality of an image displayed through a display device is remarkably degraded.